AI Paper Atlas

Research map / Detection and segmentation / CNN features

Semantic Segmentation by Early Region Proxy

Yifan Zhang, Bo Pang, Cewu Lu

CVPR 2022 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 28 citations (OpenAlex)

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace regular patch-grid prediction with learnable region proxies and per-region classificationViT-Ti/16 · ADE20K valmIoU (SS): 39.0 -> 42.1Improved
Replace regular patch-grid prediction with learnable region proxies and per-region classificationViT-S/16 · Cityscapes valmIoU (SS): 76.1 -> 75.9Mixed
Use 3–5 early Transformer layers for token/region feature extraction rather than M=0 (ablation)RegProxy-S/16 · ADE20KmIoU: 46.3 -> 47.2Improved
Use 3–5 early Transformer layers for token/region feature extraction rather than M=0 (ablation)RegProxy-S/16 · CityscapesmIoU: 76.8 -> 79.0Improved
Make the token head too deep (M=12), leaving no later layers for region-context encoding (ablation)RegProxy-S/16 · ADE20KmIoU: 47.2 -> 45.7Hurt
Increase output resolution from 1×1 to 4×4 subcells (ablation)RegProxy-Ti/16 · ADE20KmIoU: 39.30 -> 40.76Improved
Increase output resolution from 1×1 to 4×4 subcells (ablation)RegProxy-Ti/16 · CityscapesmIoU: 73.11 -> 75.08Improved
Add a 3×3 depth-wise convolution to the affinity head (ablation)RegProxy-Ti/16 · ADE20KmIoU: 40.01 -> 40.72Improved
Add a 3×3 depth-wise convolution to the affinity head (ablation)RegProxy-Ti/16 · CityscapesmIoU: 75.04 -> 74.96No clear effect
Initialize with ImageNet-21k pretraining rather than random initialization (ablation)RegProxy-Ti/16 · ADE20KmIoU: 13.96 -> 40.76Improved
Initialize with ImageNet-21k pretraining rather than random initialization (ablation)RegProxy-S/16 · CityscapesmIoU: 49.36 -> 78.71Improved
Add explicit region-semantic homogeneity regularization (ablation)RegProxy-Ti/16 · ADE20KmIoU: 40.76 -> 40.66Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy